ABSTRACT

Title of Document:

DESIGN AND APPLICATION OF PAN AND

TILT SERVO GIMBALS IN POINTING,
ACQUISITION, AND TRACKING
John Robertson Rzasa
Master of Science in Electrical Engineering
2007

Directed By:

Professor Christopher C. Davis

Electrical and Computer Engineering

Directional wireless communications systems are fast becoming an essential part of the
worlds broadband network infrastructure. When using these types of transceivers in
reconfigurable networks, it becomes necessary to point them rapidly and accurately to
different locations, or even to targets that may be in motion. The most efficient way of
doing this is through the use of two-axis pan and tilt motion stages, also known as
gimbals. This paper presents the motivation for, design and construction of, and testing
of a pair of multipurpose servo gimbals, usable for both RF and laser transceivers. The
gimbals are tested in terms of pointing error, movement speed, and response time. For
the network portion, relink times as a function of angular rotation are examined, as well
as the angular offset vs. data rate. The gimbal is also tested as part of a remote
surveillance network, evaluating its ability to track moving objects.

DESIGN AND APPLICATION OF PAN AND TILT SERVO GIMBALS IN POINTING,

ACQUISITION, AND TRACKING

By
John Robertson Rzasa

Thesis submitted to the Faculty of the Graduate School of the

University of Maryland, College Park, in partial fulfillment
of the requirements for the degree of
Master of Science
2007

Advisory Committee:
Professor Christopher C. Davis, Chair
Professor Gilmer L. Blankenship
Professor Julius Goldhar

 Copyright by
John Robertson Rzasa
2007

Dedication
To my parents and brother, to whom I owe all.

ii

Acknowledgements
After working in the Maryland Optics Group for so long, the list of people Im
indebted to could go on for ages. I owe so much to Professor Chris Davis, who gave
me a job as a sophomore undergrad and then let me continue as a grad student. It is
from him that I learned what it is like to be a true engineer, not to mention a cheerful
one! I owe much to Dr. Milner, Dr. Balzano, and Dr. Hodzic for their advice and
guidance over the years, and some good food as well! I will never forget the endless
stream of help from Dr. Shawn Ho, who formed a great engineering tag-team with
me. Ill always appreciate Dr. Sugi Trisnos patient advice, as well as Yuju Hungs
entertaining conversations. I am also deeply indebted to John J Pyle, the IREAP
machine supervisor, who taught me how to be a machinist, something Ive grown to
love very much. Finally, I thank everyone in MOG and the school in general, whove
helped me along the way and made this a fun place to work.

iii

Table of Contents
Dedication ..................................................................................................................... ii
Acknowledgements...................................................................................................... iii
Table of Contents......................................................................................................... iv
List of Tables ................................................................................................................ v
List of Figures .............................................................................................................. vi
Chapter 1: Introduction and Background..................................................................... 1
1.1
Introduction................................................................................................... 1
1.2 Mechanical Overview of Servo Motors............................................................ 3
1.3
Radial vs. Axial Flux Servo Motors ............................................................. 5
1.4
Electrical Overview of Servo Motors ........................................................... 6
1.5
Servo Motor Control Methods .................................................................... 10
1.5.1
Proportional-Integral-Differential (PID) Control ............................... 12
1.5.2
Proportional-Integral-Velocity (PIV) Control .................................... 13
1.5.3
Feed Forward Control ......................................................................... 15
1.6
Servos vs. steppers in PAT applications ..................................................... 16
1.7
Motor Selection........................................................................................... 17
1.8 Controller Selection .......................................................................................... 18
Chapter 2: Experimental Design................................................................................ 20
2.1
Design Considerations ................................................................................ 21
2.2
Design specifications .................................................................................. 23
2.3
PIV Control Loop Gains ............................................................................. 23
2.4 Motor Control through RS232 ........................................................................ 26
2.5
Mechanical Design...................................................................................... 27
Chapter 3: Experimental Results ............................................................................... 32
3.1
Introduction................................................................................................. 32
3.2
Angular Rotation Time and Positioning Error............................................ 33
3.3
Reconfigurable Network Experiments........................................................ 35
3.3.1
Transceiver Angular Offset vs. Packet Loss....................................... 35
3.3.2
Make-Break-Make Tests..................................................................... 37
3.3.2.1 RAD RF Antenna Results............................................................... 39
3.3.2.2 Canobeam Results........................................................................... 42
3.4
Gross Motion Stabilization ......................................................................... 44
3.5
Surveillance Applications ........................................................................... 47
Chapter 4: Future Work ............................................................................................. 49
Appendix A: Gimbal CAD Design Drawings ........................................................... 52
Bibliography ............................................................................................................... 55

iv

List of Tables
Table 1: Comparison of servo and stepper motors.17
Table 2: Specifications of the Bodine E-Torq Motor.18
Table 3: Gimbal Design criteria.23
Table 4: Loop gains and their associated effects25

List of Figures
Figure 1.1: Cutaway of a DC brushless servo [1]......................................................... 5
Figure 1.2: Radial and Axial flux configurations for servo motors.............................. 6
Figure 1.3: Hall-Effect Sensor [2] ................................................................................ 7
Figure 1.4: Hall-effect Sensor Alignment vs. motor current phase [1] ........................ 8
Figure 1.5: Rotary optical encoder [4].......................................................................... 9
Figure 1.6: Limit cycling behavior in a servo motor .................................................. 11
Figure 1.7: Typical servo move profile (current, velocity and position) .................... 11
Figure 1.8 PID control loop diagram [6] .................................................................... 13
Figure 1.9: PIV control loop [6] ................................................................................. 14
Figure 1.10: Copley Xenus Control Method [7]......................................................... 14
Figure 1.11 Feed Forward Control [7]........................................................................ 16
Figure 1.12: Copley Xenus Servo Drive [8] ............................................................... 19
Figure 2.1: Velocity loop filter [7].............................................................................. 21
Figure 2.2: Rotation with and without overshoot ....................................................... 22
Figure 2.3: PIV tuning method [11]............................................................................ 24
Figure 2.4: Adjustable Friction Bushing.................................................................... 29
Figure 2.5: Conceptual Design of gimbal holding a Canobeam................................. 30
Figure 2.6: Finished Canobeam Gimbal ..................................................................... 30
Figure 2.7: Conceptual design of gimbal holding a RF antenna................................. 31
Figure 2.8: Finished RAD antenna gimbal ................................................................. 31
Figure 3.1: Angular move vs. time, azimuth motor on left, elevation on right .......... 33

vi

Figure 3.2: Standard Deviation of the difference between commanded and steady
state angular rotations, azimuth motor........................................................................ 34
Figure 3.3: Block diagram of experimental setup....................................................... 35
Figure 3.4: Angular offset vs. average data rate (RAD) ............................................. 36
Figure 3.5: Angular offset vs. average data rate (Canobeam) ................................... 36
Figure 3.6: Angular move vs. Reconnect time ........................................................... 40
Figure 3.7: Time vs. data rate and motor rotation for a 180 move............................ 40
Figure 3.8: Angular move vs. Data Rate..................................................................... 42
Figure 3.9: Time vs. data rate and motor rotation for a 180 move............................ 43
Figure 3.10: Azimuthal/Elevation motor velocity and gyro velocity ......................... 45
Figure 3.11: Residue of Motor and Gyro Angular Velocities.45
Figure 3.12: Still capture from autonomous tracking experiment .............................. 48
Figure 4.1: Conceptual design of light-weight RF servo gimbal................................ 51
Figure A.1: Unloaded Gimbal Top View .................................................................. 52
Figure A.2: Gimbal side view..................................................................................... 53
Figure A.3: Gimbal Front View................................................................................. 54
Bibliography ............................................................................................................... 55

vii

Chapter 1: Introduction and Background

1.1

Introduction
In the field of directional wireless communications, being able to point a

transceiver precisely is the most important aspect in establishing a link. When the two
ends are fixed, this becomes a relatively simple process of coarsely aligning the two units
by eye or beacon, and then using some type of signal strength indicator to fine-tune the
transceivers orientations. However, this situation becomes markedly more difficult
when one has multiple nodes that must be reconfigured automatically, or if one or more
of the nodes are moving. While these situations may seem far-fetched at first, there are
many areas where they can be found, from networks between skyscraper rooftops,
between first-responder vehicles and a command center, between military vehicles and
planes, or amongst ships at sea. Knowing how to acquire, point to, and track these nodes
is a very complicated process, which has been investigated by other members of our
group, so this paper will touch only lightly on the theory involved. Once one knows
where to point, the issue becomes how to move the transceiver as fast and as accurately
as possible to the new location, or in moving cases, to track the moving node closely.
With directional wireless units regularly exceeding 100 Mbps, even a few seconds long
break in a link can cause a large loss of data. Therefore, it is essential to develop a
platform that can rotate these units to anywhere over a hemisphere or more, while at the
same time doing so reliably and quickly enough to minimize the mechanical delay
reconnect times. The most common method is to employ a two-axis pan and tilt gimbal,

which utilizes two motors to provide azimuthal and elevation rotation. Starting from this
simple idea the variations become enormous. From the gimbal on a toy to the half
million dollar one on a helicopter, there is a large range of technology used. This paper
focuses on the use of a relatively new type of motor, namely the direct-drive low-profile
(pancake) DC brushless servo motor. Amongst other things, these motors provide very
high peak torque, smooth low velocity control, as well as milli-radian positioning.
Chapter 1 of this paper details the mechanical and electrical theory behind the motors, the
differences between servo and stepper motors, and the types of controllers currently in
use. Chapter 2 explains the design and fabrication of the gimbal used in this paper, as
well as the tuning of the motor controllers. Also included are the gimbals limitations
and general specifications. Chapter 3 lists the experimental results which tested the units
positioning performance, response time, and ability to reconfigure a network link using
two types of directional wireless transceivers. Chapter 4 brings the paper to a conclusion
by suggesting future work and other potential areas of interest.

1.2 Mechanical Overview of Servo Motors

Servo motors have several distinct characteristics that separate them from their
stepper counterparts. The biggest is the lack of direct gearing between the rotor and the
output shaft. This eliminates the backlash and cogging behaviors found in steppers,
where there is a period of slop between the gear teeth before movement actually begins,
and where the shaft continues to move after the motor has stopped. This can lead to jerky
starts and stops, as well as a time delay in movement. This does not impede static
positioning performance markedly, but it presents major issues when on-the-fly velocity
changes or hard starts/stops are needed.
A model of a typical radial brushless DC servo motor is shown below in figure
1.1. For a long time, servo motors used brushes to transfer current from the static
winding to the rotor, but this would lead to wear on the brushes, in turn shortening the
lifespan of the motor. With the advent of electronic motor controllers, the brushless
design was adopted, which uses control electronics to vary the currents phases to the
motors windings in the same way the brushes do. For the rest of this paper, all mention
of servo motors will be of the brushless type.
Looking at figure 1.1 below, there are several objects of interest. First are the
armature windings (held by the stator), which create a magnetic field that travels through
the air gap to the permanent magnets on the rotor. Even though there are normally no
gears in a servo motor, cogging can still exist, as there are gaps between the magnets on
the rotor where the flux decreases, though this only becomes noticeable at low speeds.
This type of cogging in servos is perhaps more accurately termed detent torque. There
are two ways to minimize this type of cogging, the most common being the addition of

some gearing to the drive shaft. This allows the motor to run at a higher speed out of its
cogging region, but does not compromise power output or precision, though it can induce
some backlash. The other way of minimizing cogging is to skew the magnets on the
rotor so that a radial line from the center of the rotor always intersects a magnet at least
once. When using a motor without gearing, it is known as a direct drive motor. This
allows for the best transfer of power to the load, and avoids any of the negative aspects of
gearing previously mentioned. A feature in newer servo motors (including the Bodine
models used in this thesis) is the use of an ironless stator, which eliminates iron
saturation, a situation where the magnetic properties of the iron limit how much current
can be applied to the windings. Inducing iron saturation too often will cause overheating
and possibly damage the windings or magnets. With an ironless stator, rotor magnet
skewing is not necessary, as the magnetic fields arent influenced by the material of the
stator. Also, since the only mechanical connection between the shaft and the body is
through the bearings, friction is very low (especially when using ball bearings).
In high torque motors such as the ones used in this thesis, the rotor actually
consists of two plates of permanent magnets sandwiching the stator, which allows for a
major increase in torque. This feature only exists in axial flux motors, due to the design
where the stator lies in between the rotors, whereas in radial flux servos, the rotor is
completely enclosed by the stator. The majority of the heat dissipated from a servo motor
comes from the stator, so its outside location aids in cooling. In fact, the main limiting
factor in the power of a servo motor is the heat capacity of the stator and the armature
windings.

Figure 1.1: Cutaway of a DC brushless servo [1]

1.3 Radial vs. Axial Flux Servo Motors

In servo motor design, there are two types of stator/rotor configurations: radial
and axial. These refer not to the mechanical configuration but to the flow of magnetic
flux. The most common type is the radial servo, which has a smaller radius but longer
length. Normally, the stator is on the outside, completely enclosing the permanent
magnet rotor. In the case of axial motors, the rotor is a thin disk, with the stator on one
side. In high torque applications, two rotors are used, essentially sandwiching the stator.

Figure 1.2 below shows the mechanical configuration of radial and axial servo motors.
Each type of motor design has its advantages, namely axial flux motors allowing for
higher peak torque, while radial motors perform more smoothly at very low velocities.

Figure 1.2: Radial and Axial flux configurations for servo motors

1.4 Electrical Overview of Servo Motors

In older brushed servo motors, the phasing of the current supplied to the rotor was
determined by the orientation of the commutator, a plate on the rotor which interacted
with the brushes to determine where the current should flow. In this manner, the motors
operation could be controlled with a few potentiometers. With this mechanical
connection removed in brushless motors, control became more complicated. All
commutation became electrically controlled, requiring the use of motor controllers that
employ some type of feedback sensor to determine the rotors position relative to the
stator. The most popular sensors in use are Hall-effect sensors, optical rotary encoders,
and resolvers.

Hall-effect sensors operate by changing their voltage in response to a varying

magnetic field. Thus, as the magnets on the rotor move, the voltage will change
proportionally. Using several sensors spaced evenly around the rotor, one can determine
the rotor position as well as its velocity. Hall-effect sensors have the added benefit of
having no mechanical connection to the rotor, and are usually integrated into the armature
windings. In most motors, they are used as the primary sensor to determine the required
current phase to send to the motor, given their fast response and lack of additional control
circuitry. A diagram of three Hall sensors in a servo motor is shown in figure 1.3, while
a chart of the Hall sensor alignment and current phase for the Bodine E-Torq motor is
shown in figure 1.4.

Figure 1.3: Hall-Effect Sensor [2]

Figure 1.4: Hall-effect Sensor Alignment vs. motor current phase [1]
In some cases, such as for stepping motors, velocity measurements are not
required, but instead position. In this instance, the optical rotary encoder is the primary
sensor, since no commutation control is needed. It uses a slotted disk attached to the
rotor with either a binary or gray code cut into it. An IR signal is passed through the code
wheel, and the resulting pulses are processed in the controller to determine the rotors
position. Of course, as the resolution goes up, so does the encoders complexity and cost.
Current encoders allow for milli-radian precision [3]. The two main types are absolute
and incremental encoders. The former measures the position based on an etched pattern

in the wheel, while the latter measures interrupt pulses through a slotted disk. Absolute
encoders are the most popular, since they can also be used in an incremental fashion.
Figure 1.5 shows a diagram of an optical encoder.

Figure 1.5: Rotary optical encoder [4]

An angular resolver uses a special rotary transformer to determine the rotors
position. The theory behind these is relatively complicated, so the reader is referred to
[5] for more detail. They can provide accuracy on par with an optical encoder, and their
signals can be passed through a slip-ring, something useful for reducing cabling.
Resolvers tend to be complicated devices requiring sophisticated control circuitry as well
as an additional rotary transformer to provide power to the resolver itself, so they do not
appear often in positioning applications, or situations with size and power constraints.

1.5

Servo Motor Control Methods

When using a servo motor in a velocity control mode, the encoders position is

used to form a velocity estimate, which in turn adjusts the underlying current loop to
achieve the desired motion. It is worth noting that since the velocity used by the
controller is an estimate, the velocity error is bounded by the resolution of the encoder.
Thus, for positioning and constant velocity applications it is essential to have a high
resolution encoder, otherwise the controller wont be able to keep the motor moving
steadily (including at zero velocity). At zero velocity, the motor could bounce between
encoder pulses, a phenomenon known as limit cycling (figure 1.6 with a frequency of
~2Hz). The primary method of eliminating limit cycling in servo motors is to introduce
some static friction into the drive train, usually with an external bearing. Most servo
motors have more than enough torque to overcome the friction of the bearing without a
noticeable loss in performance.
Many servo controllers use a trapezoidal velocity profile to move to set positions,
with a linear acceleration phase, a constant velocity phase, and a linear deceleration
phase, hence the title trapezoidal. The control loop must be tightly tuned for this to
occur properly, since any mistake can cause the motor to overshoot its target, move
unsteadily during the constant velocity portion, arrive too slow, or even oscillate once
reaching the position. A typical trapezoidal profile is shown in figure 1.7, comparing
commanded and actual current, velocity, and position.

10

Figure 1.6: Limit cycling behavior in a servo motor

Figure 1.7: Typical servo move profile (current, velocity and position)

11

To ensure that a motor is moving in the proper fashion, some type of feedback control
loop must be used, with the rotary optical encoder as the feedback source There are two
types of these loops used in motor control, the first (PID) being common to many
different types of control, not just motors. The second (PIV) is used almost exclusively
in motor control, velocity being the most important factor.

1.5.1 Proportional-Integral-Differential (PID) Control

In a PID control algorithm (figure 1.8), the system operates to minimize the error
between the output signal, (s), and the commanded signal, *(s). Each of the three
parts (proportional, integral, and differential) play an important part in the control
process. The proportional gain Kp adjusts the overall level of the output signal, driving it
closer to the ideal output. However, this signal is rather slow and also has the
disadvantage of never being able to drive the error signal to zero. Thus, if Kp is set to
high, the output signal will never reach a steady state in static operations, such as a motor
in stand still. To compensate for this, the integral factor Ki is introduced. This takes a
running integral of the error over a set period of time and feeds it back into the output
signal, helping to zero the steady state error. Using this, one can achieve a steady state
signal with no error, as would be expected in a still motor. Because the integral term
adjusts for error accumulated over time, it cannot always eliminate rapidly occurring
error. There are two ways to compensate for this, namely adding a differentiation
component to the error signal, or using feed forward control, which will be described in
section 1.3.3. Adding the differentiation term, Kd, allows for a quicker control loop,
since it contributes the rate of change of the error to the output signal. One must
12

carefully use this term to avoid introducing high frequency noise. The most common
way of eliminating this noise is to add a low pass filter.

Figure 1.8 PID control loop diagram [6]

1.5.2 Proportional-Integral-Velocity (PIV) Control

A derivative of the PID controller is the PIV controller, which uses the velocity of
a sensor as part of the error correction algorithm. This is especially useful in the field of
servo motors, since controlling velocity is extremely important (eg. Electric trains,
industrial automation) and can be easily measured using optical encoders or tachometers.
A formulaic diagram of a generic PIV controller is shown in figure 1.9. A diagram of the
related control loop used in the Copley Xenus drive is shown below in figure 1.10.
Looking at figure 1.9, we see two new gain factors, Kv and Kt, Kt being the integral gain.
Kv is known at the velocity proportional gain, and allows the control loop to react
quickly to velocity changes. That is, when one wants to accelerate or decelerate a motor
rapidly, Kv should be high. However, leaving this value high for slow movements will
cause overshoot. In situations where a steady velocity is required, Ki should be
increased, as it will attempt to reduce the velocity error over time.

13

Figure 1.9: PIV control loop [6]

Figure 1.10: Copley Xenus Control Method [7]

14

1.5.3 Feed Forward Control

In some cases it may be necessary to eliminate following errors or overshoot in
servo control systems. In some applications, it doesnt matter if the output control signal
lags behind the commanded input, or if the motor overshoots a little before coming to a
stop. However, in precision positioning or velocity matching applications, these two
issues can become big problems. When a positioning command is issued, the goal is for
the gimbal to move to that location as fast as possible and then come to a dead stop. If
there is overshoot, this could cause the data transceiver to lose its connection before relinking successfully, or a camera to initially miss its target. This would also potentially
cause the network topology control to malfunction, if it sees nodes intermittently
connected. To induce faster response times in the control loop, the most common method
is that of feed forward control, which directly couples the commanded velocity and
acceleration to the control loop. The main requirement of feedforward control is the
availability of acceleration and velocity commands from the motor controller, something
usually found only in new units. Use of feed forward control will minimize the tracking
error, which is the difference between the commanded and actual velocity. If this error is
low, the effective response time of the control loop becomes faster, which in turn
decreases the chance of overshoot. When tuning a motor using feed forward control, one
adjusts the velocity feed forward gain (Vff) first to achieve the desired overall response
time, that is, Vff is usually set as the value that causes about 10% overshoot. This small
overhead is needed because the acceleration feed forward will decrease the response time
of the velocity feed forward somewhat. Finally, the acceleration feed forward gain (Aff)
is adjusted in order to eliminate any overshoot caused by Vff [8]. In the end, feed

15

forward predicts the motion based off of a predetermined profile, while P.I.V adjusted for
unexpected disturbances to that profile. A diagram of a PIV control loop using feed
forward is shown in figure 1.11.

Figure 1.11 Feed Forward Control [7]

1.6 Servos vs. steppers in PAT applications

In our groups research, gimbals are utilized in pointing, acquisition, and tracking
applications. For different tasks, different motors are needed. When developing the
homography and radial trifocal tensor algorithms for image-based acquisition and
pointing, Dr. T-H Ho used a gimbal with micro-stepping motors that have a 0.0072 step
resolution. This type of stepper gimbal can point to a location with great accuracy and
repeatability, but cannot track moving objects nor change direction mid-flight. In the
experiment described later in section 3.5, a servo motor was equipped with a camera to
track moving vehicles from a remote location. While it used the same algorithms as in
[9], its pointing accuracy was only about 20% as good compared to the stepper motor

16

gimbal. However, for aiming a camera at large objects such as cars, this did not become
an issue. What was more important was its ability to match velocity with an object so
that it was always in the FOV of the camera, something which stepper motors cannot do
smoothly. A comparison of servos vs. steppers (of similar continuous torque ratings) is
shown below in table 1.

Brushless Servo Motor [1]

Micro Stepping Motor[4]
2.3 Nm
3.5 Nm
>23 Nm
8.3 Nm
8,192
50,000
590W (continuous use)
380W (continuous use)
None (direct-drive)
Harmonic (50:1)
Optical encoder and HallOptical encoder
effect sensors
Operating modes
Current, velocity, position
Velocity, position
Table 1: Comparison of servo and stepper motors
Continuous torque
Peak torque
Pulses per revolution
Power consumption
Gearing
Feedback

1.7

Motor Selection
Before deciding what motors to use for the various planned experiments, it was

important to select the appropriate qualities needed. First and foremost is the motors
ability to function at slow (<30rpm) velocities with a small tracking error. After this
came the requirement of being able to change velocity and/or final position while already
in motion. In other words, if a motor is traveling to point A, the controller can send an
updated movement profile mid-flight to redirect the motor. In the case of a stepper
motor, one must wait until one motion is finished before starting another. Even with fast
acceleration and deceleration, this command delay causes jerky motion in stepper motors
when used in tracking applications.

17

Finally, the azimuthal motor must be able to swing a mass of 30kg 180 in under
two seconds. This requires a large peak current value, which in experiments was around
35ADC.
After all these considerations were taken into account, the motor selected was a 7
diameter DC brushless servo motor from Bodine-Electric, model number 07-EKEP-00.
A small list of relevant specifications is shown below in table 2.
Maximum Continuous Current
Motor Power
Maximum speed
Maximum continuous speed
Maximum continuous torque
Peak torque
Motor efficiency
Number of poles
Angular resolution (8192 ppr)

5.5 ADC
590W
6000 rpm
2500 rpm
2.3 Nm
>23 Nm
88%
8
0.0439

Table 2. Specifications of the Bodine E-Torq Motor (adapted from [1])

1.8 Controller Selection

There are a myriad of servo motor controllers on the market today, some of the
most popular being from AMC, Copley, Parker Motion, and Oriental Motor. All provide
the same basic function: supplying three-phase DC current to the motor in the proper
pattern to induce the desired motion in the motor. Most controllers offer different control
modes, feedback options, and output current levels, so they are generally chosen for a
specific application. In our case, the controllers for this project needed to provide at least
these five features: High peak current, optical encoder with milli-radian resolution, feed
forward control, trajectory position mode, and the ability to control the motors and gains
in real-time through RS232 or CAN.

18

From these considerations and the recommendations of the motor distributor, the
Copley Xenus driver was chosen, as it fulfilled the minimum requirements for our project
in addition to its affordability. Figure 1.12 shows the Xenus drive, the XSL-230-36

Figure 1.12: Copley Xenus Servo Drive [8]

One interesting safety feature of this driver is its I2T algorithm [8], which prevents
too much current being supplied to the motor. This algorithm averages the power over a
certain time interval, which for our setup was 1.1s. It constantly monitors the motors
actual current draw and checks whether the motor is operating above its continuous
current limit. If it is, it increases an internal accumulator, and if it not, the accumulator is
decreased, though never becoming negative. If the accumulator goes higher than a
certain value determined by the time interval, continuous current limit, and peak current
limit, the controller shuts the motor off. The I2T time constant for this project was set
intentionally low to protect the motors and drivers.

19

Chapter 2: Experimental Design

The central piece of hardware in this paper is a two-axis gimbal which can carry
either a directional RF antenna or FSO optical transceiver. The FSO transceiver (a
Canon Canobeam DT-110) weighs 8kg and is significantly larger than the RF antenna, so
the gimbal was designed with this unit in mind. Most gimbals follow the same design
rules, with one motor providing azimuthal rotation and the other elevation rotation. A
payload cradle is coupled between the elevation motor and high-speed ball bearing to
hold the transceiver. Because the elevation motor contributes a great deal to the moment
of inertia of the entire upper section, a counterweight is used to balance the mass of the
elevation motor, which sits far out from the center of azimuthal rotation. While the
counterweight also adds to the moment of inertia, it helps the motor maintain constant
velocity as well as smooth acceleration and deceleration, as the moment of inertia is now
balanced. It is worth noting that this is a prototype, not having the great advantage of
slip-ring connectors to transfer power and control signals to the elevation motor and
payload without the need for cables, which can add uneven forces to the elevation motor,
in turn causing errors in the control loop.

20

2.1 Design Considerations

When designing a gimbal, one must take several issues into consideration. First is
weight; namely, the maximum payload mass which will determine the maximum
acceleration and deceleration values that can be achieved without inducing instability.
This worst-case approach works well when tuning servo motors, as the controller can
easily provide less current for a smaller moment of inertia. If one tunes the motor for a
lower inertial payload and then puts something heavier on, the motors will perform
unevenly, and in some cases even cause the controller to shut off.
The second consideration is that of the maximum velocity. For a device that
rotates always at a constant velocity, how well it accelerates is not really an issue. All that
matters with a conveyor belt is that it moves steadily, not how fast it spins up in the
morning. However, when the motor makes rapid starts and stops, velocity stability is
essential in creating a smooth motion profile. Therefore, the velocity feedback loop takes
prominence in the design. A diagram of the velocity loop used in the Xenus amplifier in
shown below in figure 2.1.

Figure 2.1: Velocity loop filter [7]

Lastly, one must decide whether stopping on a dime is more important than speed.
With such a large moment of inertia, it is quite difficult to make a gimbal stop suddenly
with no overshoot. This can be eliminated by moving slower, but this can become

21

undesirable in certain situations that will be discussed in chapter 3. However, in practical

situations seen in experiments with directional RF antennas, some overshoot can be
tolerated, as the beam is wide enough to still cover the receiver even if there is some
misalignment for a short period. Figure 2.2 below shows a comparison of movement
profiles of the same step count, part a) moving fast enough to cause overshoot, while b)
moves slower.

Figure 2.2: Rotation with and without overshoot

Naturally, one must also think about things like power, reliability, rigidity, weight,
cabling vs. sliprings, and many other issues when designing such a complicated system.
Since this is a prototype to test various PAT concepts, the above concerns will be left to
the next generation of gimbals.

22

2.2 Design specifications

As mentioned briefly in Chapter 1, the specifications for this gimbal are based on
its usage as a platform for rotating a directional transceiver. Another factor was the
ability for the gimbal to be used as a platform to compensate for large platform motions,
say from the rocking of a ship. This way, an FSO transceiver could use rapid fine-angle
pointing, acquisition, and tracking to lock on and follow another transceiver. A prototype
FPAT system utilizing piezo actuators has already been designed and tested in [9].
Table 3 below shows the target specifications for the gimbal.
Time to rotate 180 azimuthally
Field of Regard
Payload Capacity
Angular resolution
Weight (with max. payload)

~1.5s
3 Steradians
10kg
<0.5
Less than 45kg

Table 3: Gimbal Design criteria

2.3 PIV Control Loop Gains

To achieve the above specifications, robust tuning of the motors controller is
required. This is a complicated process, exacerbated by the fact any change in the
weight, size, or moment of inertia will necessitate a retuning. However, there is a wellestablished method for tuning the PIV controller, which is of the form current loop =>
velocity loop => position loop. In total, there are eight gain factors that must be adjusted,
and also the velocity filter cutoff frequency. Figure 2.3 shows a standard flow chart for
tuning the velocity loop of a PIV controller [10]. After one has tuned the velocity loop,
the position loop can be tuned, usually by commanding a step movement, viewing the
motor response, and adjusting the position proportional gain, feed forward gains and gain
multiplier[11].
23

Figure 2.3: PIV tuning method [11]

After this process has been completed, the cutoff frequency of the velocity loop
filter must be adjusted. It essentially tells the motor feedback to ignore certain amounts
of noise. If this value is set too high, the controller will try to compensate for motion that
is in fact not there. When this happens, the motor will begin to oscillate or make a
grinding noise as it rapidly changes current polarity in an attempt to compensate for the
perceived noise. In our system, the appropriate value was found to be 3Hz, using a
single-pole 20db/dec low-pass filter.
With regard to the other eight loop gains, table four below explains the effect each
of them has on the system.

24

Gain Type
Current Proportional
Current Integral

Associated Effect
Allows for faster current changes
Smoothes output current, reduces steady
state error
Velocity Proportional
Allows for faster velocity changes
Velocity Integral
Damps out velocity overshoot, reduces
steady state error
Position proportional
Determines how fast motor tries to get to
a position, too high a value causes
overshoot
Velocity feed forward
Decreases following error during
constant velocity moves
Acceleration feed forward
Reduces overshoot caused by using
Velocity Feed Forward
Gain multiplier
Affects overall response time of the
position loop. Too high a value causes
oscillations when settling
Table 4: Loop gains and their associated effects
When one tunes a motor, it assumes the motor will only be used in that way (same
move distance, speed, acceleration, etc). Ideally, the underlying current loop should be
controlled by a DSP which actively monitors the motors position, velocity, and
acceleration, and then dynamically adjusts the current. This is essentially the same as
adjusting the gain values continuously. However, for the types of movements described
in this paper, the two main gains that need to be adjusted are Vp and Pp, which are
changed adaptively through serial port control based on the acceleration value selected.
To find the best values, movements were commanded using various acceleration values,
and Vp and Pp were adjusted to achieve the best performance. These values were then
curve-fitted to a linear slope, which was in turn used as a calibration for adjusting the
gains for any other motion. The velocity value was set based on the rotation angle, in
order to ensure a triangular profile (i.e, always accelerating or decelerating). For the

25

surveillance applications described later in 3.5, the gains and velocities were adjusted
dynamically, but were limited to 10Hz by the motion detection program.

2.4 Motor Control through RS232

While there are several different control schemes available for the controllers
used in this project, the simplest to employ rapidly is the RS232 serial command
interface. Through a series of ASCII commands sent to the controller, the user can
operate the motors in current, velocity, or position modes, as well as being able to change
the loop gains. In some ways it mimics what a DSP would do, but at a far slower rate.
For these experiments, all control came from a custom-designed Labview interface, using
the ASCII commands from Copley. In experiments, movement commands could be sent
at about 20Hz. This is adequate for the surveillance tracking applications described later,
but it is not enough for rapid stabilization experiments. There is also an inherent serial
communications delay of 5ms, so in the case of the gyro test in section 3.5, a delay can be
seen between the gyro sensors and the motors movement. For a general positioning
application, figure 2.4 shows the command sequence to complete a motor movement.
The section boxed in blue must be resent for each new command, assuming the values for
velocity, acceleration/deceleration, and move distance have changed.

26

Set Motor to
Trajectory Mode

Enable Motor

Set Acceleration

Set Deceleration

Start Move

Set Loop Gains

Set Move Count

in Steps

Set Velocity

Figure 2.4: Command sequence for a position-trajectory move

2.5 Mechanical Design

Most gimbals in use today use a basic U-shaped structure, with the azimuthal
motor underneath, and the payload cradle supported by the elevation motor and an
opposing bearing on top. There are many small variations on this, but the authors design
was simple enough to be made in the IREAP machine shop. The motor controllers were
attached to the azimuthal motor support to allow for short cabling, and to make the unit
as compact as possible. The overall dimensions of the finished FSO gimbal are
71.12x48.26x45.72cm. Weight reduction was achieved by using mostly aluminum, and
slotting out parts of the structure. A brass counterweight was added opposite the elevation
motor in order to provide a balanced moment of inertia. Even though this added mass
slightly decreased the maximum velocity of the azimuthal motor, the balanced load
greatly improved positioning performance and velocity stability, not to mention
simplifying the tuning of the PIV loop. The motors were operated in a direct drive
configuration, due to the large moment of inertia involved. While using gearing could
give even better low-speed performance, the gears would have to be specially designed to
handle the large peak torques generated by the azimuthal motor. To eliminate limit

27

cycling, a sealed ball bearing was used for the elevation stage, while a custom copperbacked bushing was used for the azimuthal motor. This design was employed because it
allowed the static friction to be adjusted in order to provide just as much friction needed
to stop the steady state oscillatory behavior. Figure 2.4 below shows a schematic of the
adjustable friction bushing. Figures 2.5 and 2.6 show the authors conceptual design of
the unit holding a FSO transceiver and RF antenna, respectively, while figures 2.7 and
2.8 show the actual units holding a FSO transceiver and RF antenna, respectively.

28

Figure 2.4: Adjustable Friction Bushing

29

Figure 2.5: Conceptual Design of gimbal holding a Canobeam

Figure 2.6: Finished Canobeam Gimbal

30

Figure 2.7: Conceptual design of gimbal holding a RF antenna

Figure 2.8: Finished RAD antenna gimbal

31

Chapter 3: Experimental Results

3.1

Introduction
After all construction was complete, the end result was a pair of gimbals, each

capable of holding either a RAD RF antenna [11] or a Canobeam FSO transceiver [12].
Before using these units to test the gimbals performance as part of a reconfigurable
wireless network, it is essential to see how they perform in basic positioning tasks,
measured in terms of movement times and positioning error. Next, the gimbals were
evaluated as communications nodes, measuring the average data rate during a makebreak-make test, which will be described later. Another experiment was also performed,
in which the gimbals capability as a gross-motion stabilization platform was evaluated.
Finally, a small demonstration of object tracking using automatic speed control was
conducted. In the latter two experiments, the main limiting factor was the use of serial
port commands to control the motors, which induced a 60ms delay (the total delay from
all the commands seen in figure 2.4). This prevented the gimbals from quickly
compensating for rapid changes in platform orientation, and also caused some
unevenness when tracking moving objects. The use of DSP would drastically reduce the
delay and command time, however the amount of development time was not feasible for
the scope of this phase of the project. In a mature system, the encoder feedback would be
directed straight into the DSP (which would also calculate the velocity and acceleration
estimates), thus eliminating the various delays incurred from interfacing with the motor
controller.

32

3.2 Angular Rotation Time and Positioning Error

The first performance test involved moving the gimbal as rapidly as possible in
both elevation and azimuthal directions, and monitoring how long the total move took. In
the case of the azimuthal motor, the movement range was tested between 0 and 180, and
-30 to 30 for the elevation, both limited by the mechanics of the system. Figure 3.1
shows the results for both motors, with each rotation angle tested ten times, with the
Canobeam as a payload. Figure 3.2 show the standard deviation of the final position,
showing how well the motors can move to a commanded position, averaged over 10 runs.
The gains for the amplifier were dynamically adjusted in this test to maximize speed and
minimize overshoot.

Figure 3.1: Angular move vs. time, azimuth motor on left, elevation on right

33

Figure 3.2: Standard Deviation of the difference between commanded and steady state
angular rotations, azimuth motor
Looking at the above graphs, a few interesting observations can be made. First,
the move times for the azimuthal motor get faster above a certain angle, due to the
momentum of the gimbal once it gets going. As for the elevation motor, the move times
increase with the rotational angle as expected, however for very small angles, it also
increases. This is primarily due to the static friction of the ball bearing on the elevation
motor assembly. Looking at the graphs of the position standard deviation, the azimuthal
motor performance remains nearly flat across the movement range, and shows a very
small (<0.05) deviation. No graph was included for the elevation motor, as the standard
deviation of the angular rotation (defined as the difference between the commanded and
steady state position) was 0 over the entire angular range. Since the elevation motor has
much less inertia to deal with, the position loop was able to converge to the command
value.
34

3.3 Reconfigurable Network Experiments

The test setup for the experiments in this section was located on the roof of the
Kim building, with the remote node on Martin Hall, 254m away. Figure 3.3 below shows
a diagram of the test bed and hardware. Two transceivers were tested, a Canon
Canobeam FSO unit [13], which uses at 7mW laser operating at 765nm and has an
effective range of 500m, and a RAD Airmux-200 Radio unit [12], which operates at
5.81GHz and has an effective range of 3 miles at its full data rate using a 30cm x 30cm
patch antenna. Both of these units are designed for fixed outdoor links, to the following
experiments also tested whether they could function in a morphing physical network.

Smartbits

Transceiver on
text
Gimbal

Remote
Transceiver
254m Link Distance

Loopback

Laptop

Figure 3.3: Block diagram of experimental setup

3.3.1 Transceiver Angular Offset vs. Packet Loss

Each of the transceivers described in this thesis has a certain angular mismatch it
can tolerate between two nodes. This is very important to quantify, as it can reveal how
high the pointing resolution must be for the motors, as well as the units actual
beamwidth. To begin with, each of transceivers was aligned manually to achieve the
strongest signal strength, based on the manufacturers alignment procedure. This

35

position was then set as the starting location. The azimuthal motor was then rotated
incrementally until the signal was lost, with a Smartbits test rack recording the average
data rate over a one minute interval at each angle. The same procedure was then repeated
for the elevation motor, with the azimuthal motor recentered. Figures 3.3 and 3.4 below
show the results for both transceivers.

Figure 3.4: Angular offset vs. average data rate (RAD), azimuthal and elevation

Figure 3.5: Angular offset vs. average data rate (Canobeam), azimuthal and elevation
We can make several observations from the graphs above. The effective
beamwidth of the RAD antenna is about 15, compared to the specification of 9, which
was based off a measurement along the antennas diagonal. These radios also display a
sharp cutoff in usable data rate; that is, there is a critical angle where the data rate goes

36

almost immediately to 0 Mb/s. Also, in order for the RAD antenna to achieve its full
capacity, all 24 sub-channels must be synchronized. As the signal strength fades, these
sub-channels start to fail, leading to an immediate drop off in data rate.
As for the graphs from the Canobeam, the first thing that can be noticed is the
flatness of the data rate in the usable region. Not surprisingly, the use of a single
wavelength to transmit data leads to a tighter received signal, and thus a steadier data
rate. The Canobeam was found to have a 2.8 horizontal and 3.1 vertical beamwidth,
with the signal fading to nothing within 0.05 past these limits. These numbers are much
wider than the <1 beamwidth specified by Canon, however the autotracking feature
was enabled in this test, which can compensate for 1.2 of angular mismatch. This is
actually a very useful feature, since it makes the initial alignment much easier. It also
acts like a stabilizer for very slow platform motions. It is feasible that this feature could
be used on a ship to compensate for some of the very low frequency oscillations (i.e. a
ship in calm waters), thus saving gimbal motor power. One interesting observation from
the next section was that the gimbal rotated fast enough so that the autotracking did not
begin searching for the beam, implying that its response time is greater than ~2.5s.

3.3.2 Make-Break-Make Tests

The end result of all these tests was to optimize the gimbals so that they could
function in a reconfigurable network. To that end, they must able to rotate to a position
fast enough that the network can relink in a time so that little data is lost. Of course,
little data is a relative term given buffering and other tricks, but for our purposes we
used the Smartbits packet generator to test the links unbuffered. The idea is to simulate

37

the gimbal linking to a node that would be the farthest away (180 from the starting
position). To do this properly would require three transceivers of each type, which we
did not have. To bypass this, we moved the gimbal up to 90 away and then 90 back to
the original position. This method will result in slower relink times due to the pair of
accelerations and decelerations, but it is reasonable in obtaining a measure of the worstcase relink time. The test methodology is as follows: packets are sent from the Smartbits
at the fixed location and then looped back at the mobile node. The link is then broken in
the above manner while the data rate is measured. This is then repeated 10 times for each
rotation angle. Since each of these units operates at different data rates and use different
modulation schemes, it would be improper to directly compare their performance. Also,
these units are designed for different areas, so a group using a FSO transceiver wouldnt
necessarily want to use a radio unit. It is therefore advisable to look at these tests
independently, given the units fundamental differences.
For both experiments, the test range was between Martin Hall and the Kim
Building, which is a distance of 254m. Tests were performed on a clear day with winds
less than 10kph. The make-break-make tests were conducted by rotating the azimuthal
motor, since in the case of the RAD antenna the vertical polarization must remain the
same, and for the Canobeam because the unit is too big to do a vertical flip, so neither of
these situations is likely to occur in real topologies.

38

3.3.2.1

RAD RF Antenna Results

For the RAD antenna units, the radios were first manually aligned to find the
maximum signal between the two. After this was set as the zero position, the unit was
rotated azimuthally in the manner prescribed above, and the data rate was recorded.
When used at its maximum signaling rate, the RAD antenna uses 24 sub-channels to
transmit and receive data. Each of these takes at most 300ms to reconnect, so if they all
do not re-sync on the first pass, the system must try again for the link to be reestablished.
Because of this, reconnects can take up to eight seconds if all the subchannels connect on
the first pass. Figure 3.5 below shows the results of the make-break-make test, plotting
total angular move (away and back) vs. reconnect time (at 15.5Mbps, slightly below the
maximum capacity of the unit, in order to allow some cushion). Figure 3.6 shows a
graph of time vs. data rate for a typical move of 180, as well as the movement of the
motor. In figure 3.6, the motor doesnt actually accelerate as sharply as shown, this was a
data logging issue caused by the program controlling the motors. The position looks like
a steep triangle if recorded properly.

39

Figure 3.6: Angular move vs. Reconnect time

Figure 3.7: Time vs. data rate and motor rotation for a 180 move
40

Looking at the previous plots, one can note several interesting properties of the
radios. First is that the relink time is almost unrelated to the time that the motor takes to
move away and back. Of course, for a longer mechanical break the reconnect time would
be longer, but these tests show that the gimbals moves fast enough at all angles that it
doesnt contribute significantly to the downtime. The manufacturer stated that the
maximum reconnect time is around 8 seconds, however in these experiments it was found
to be around 11.45 seconds. This disparity is most likely from the sub-channels not all
disconnecting and then reconnecting at the same time, somewhat akin to pulling taffy
apart. If the radio cannot reconnect all the sub-channels at once, it rescans them all again,
which leads to a longer overall reconnect time. Given that the make-break-make times
for the motor ranged from 1.1s (10 rotation) to 2.9s (180 rotation), the gimbal
movement represents between 9.6% and 25.3% of the link reconnect time. It should also
be noted that for rotations less than 10, the radio did not lose its connection at all. The
use of lower frequency directional RF transceivers offers a significant saving in terms of
cost and alignment difficulty, but at the cost of reconnect time. However, these units
were never intended to be used in situations involving rapid reconnects, so slow relinking times are somewhat to be expected. Utilizing a higher frequency radios would
speed up reconnect times significantly, as they only use two channels, one for Rx and one
for Tx.

41

3.3.2.2

Canobeam Results

While the Canobeam doesnt have the multi-channel connection issues of the
RAD unit, it does have a much tighter beam, which can make any motor overshoot or
beam misalignment a problem. However, when swinging such a large mass around, there
is a definite tradeoff between speed and settling time. Figure 3.7 below shows the results
of the make-break-make test, plotting angular move vs. reconnect time (at 95 Mbps, near
the maximum single channel capacity of the unit). Figure 3.8 shows a graph of time vs.
data rate for a typical move of 180.

Figure 3.8: Angular move vs. reconnect time

42

Figure 3.9: Time vs. data rate and motor rotation for a 180 move

Looking at the results from the Canobeam, we can see several striking differences
from those of the RAD antenna. First, is that the reconnect times do increase as a
function of the angular rotation before leveling off. Because the reconnect time of the
Canobeam itself is relatively short (~1.5s), the mechanical rotation time does become a
major factor in the overall reconnect time. This can be seen in figure 3.8, which shows
that the gimbals movement for a 180 break accounts for about half of the links
downtime. Thus, in applications involving the reconfiguration of FSO nodes, the
mechanical rotation time is a dominant factor in reconnect delay. When using FSO
transceivers, the received data rate is very sensitive to platform jitter, and because of the
lack of buffering inside the unit, the data rate can drop dramatically for short periods of
time when the platform moves by even less than a degree.

43

3.4 Gross Motion Stabilization

Another part of this project was the requirement that these gimbals could operate
as stabilization platforms, in addition to being transceiver pointing devices. If, for
example, two ships wished to communicate using directional wireless transceivers, the
rocking of the ships would need to be compensated for in order to give the transceivers a
stable platform to work from. In modern aircraft, a gyroscope is used to collect angular
velocity measurements and then adjust the control surfaces appropriately to produce a
stable airplane. In this experiment, a 6 DOF gyro from O-Navi was used, with 300 /s
angular velocity sensitivity [13]. The basic premise for these tests was how well the
motors could match a commanded waveform, which is essentially what it would have to
do to compensate for platform motion.
The test was a somewhat realistic setup, where the gimbal had to move in the
opposite velocity to a quasi-random input waveform, courtesy of the previously
mentioned gyro. The gyro was moved by hand, and the gimbal had to compensate for it.
The ultimate test would be to mount the gimbal and gyro on either a hydraulic motion
platform or a boat, which would provide adequate large-amplitude motions for the system
to compensate for. The signals from the elevation motor and gyro Y sensor may look
jerkier than those of the azimuthal motor, however both have the same amount of jitter,
the graphs are merely scaled differently to show the movements better.

44

Figure 3.10: Azimuthal/Elevation motor velocity and gyro velocity

45

Figure 3.11: Residue of Motor and Gyro Angular Velocities, Azimuth on left, Elevation
on right
The residues of the velocities (defined as the sum of the motor and gyro
velocities) show that both motors are able to match the gyros output well, with the
elevation motor performing exceptionally well, due to its lighter payload. Some of the
error seen above arose from uneven sampling due to the serial port communications
delay. As mentioned earlier, the average delay for each command sequence is 60ms, but
this can vary by 10ms. The large spike in the residue of the azimuthal motor was a
result of jerking the gyro to its sensors limit, but the motor was not able to accelerate
quickly enough with such a large load attached. However, it is unlikely that a
stabilization platform would have to accelerate to 150/s over 600ms.
From this small test of the motors and controllers, it is apparent that they have the
capacity to function as part of a stabilization platform. The use of an analog gyroscope
with its output connected directly to a DSP controlling the motors would form the basis
for a high performance stabilized platform. Preliminary tests from a colleague have
shown the communications delay from a similar system using the CAN bus to be around
1ms, so digital control is still possible in the same fashion as the RS232 here.

46

3.5 Surveillance Applications

Another area of research in the Maryland Optics Group is autonomous remote
real-time surveillance systems. With this, a camera can analyze a scene and determine if
motion is present, whether the object is a person, car, truck, etc..., if an object was
abandoned, and all kinds of other processes. Currently, a HDTV camera is mounted on a
stepper motor gimbal to provide zoomed-in follow-up on events. However, these motors
cannot provide smooth tracking of moving objects, which leads to blurry video and the
inability to get clean image data. Stepper motors generate a current profile before
actually moving, so when in motion the movement cannot be changed. In contrast,
certain servo controllers can operate in a trajectory position mode [7]. This mode also
creates a current profile to move the motor, but it can be changed at any time during the
move, which, when tracking moving objects, allows the motor to act like its in velocity
mode, but keep its precise positioning capability. In this experiment, one of the servo
gimbals described in this paper was fitted with a HDTV camera and told to automatically
follow the first person that entered the FOV of a fixed camera next to the gimbal. A still
image from the video is shown in figure 3.12 as an example of the image quality
available. A video of the gimbal in action can be seen at [15]. This application is
particularly encouraging because of the potential for multiple camera-mounted gimbals to
track the same object over a long range, handing off from camera to camera, regardless of
the movement behavior of the object. While this has yet to be fully implemented, it is of
high interest to the campus police department.

47

Figure 3.12: Still capture from autonomous tracking experiment

Looking at the above screen capture from [15], the image is quite clear, even
though it is from a 24fps camera outputting compressed video while the camera itself is
in motion. Several license plates can be readily identified from >50m away without
image enhancement. In viewing the video, we see that the gimbal tracks the person
smoothly, with the exception of a few spots. The motor receives movement commands
from the motion detection program, which will take longer between frames when there
are more objects in the scene, so the motor will then receive unevenly spaced commands,
leading to changes in the velocity. This can be mitigated by intentionally slowing down
the motion detection program to a period that encompasses most detection times. The
gimbal will then move more smoothly, but at the expense of being able to follow objects
that are rapidly changing direction. However, cars and people generally do not change
direction very rapidly or move so far that they would fall out of the FOV of the camera,
given the motion detection programs update rate (~10fps).
48

Chapter 4: Future Work

While the experiments and devices described in this thesis are varied, there is
much left to be worked on. Since they are only prototypes, certain areas were not
developed because of time and cost constraints. The system described in this thesis has
the definite potential to become a robust centerpiece of a reconfigurable wireless
network, provided several improvements are made.
One of the first improvements would be the use of a higher resolution optical
encoder which would increase the precision of positioning moves. Encoders of up to
20,000ppr are available, but again at a higher cost. While not necessary for RF
applications, these would be essential for positioning FSO transceivers at long distances,
or for stabilizing rapidly moving platforms. The use of higher resolution encoders can
also mitigate limit cycling, where the motor oscillates between encoder pulses while at a
stand still. Since the pancake motors used in this paper are not mass-produced units,
upgrades like the ones above will add time to the production of a motor, the ones we used
taking five weeks to construct.
Another upgrade that would be very time consuming, but not very expensive,
would be controlling the motors using a DSP chip. The Xenus controllers currently allow
for full motor control using analog voltage inputs. While not critical for discrete
positioning, the use of a DSP would vastly improve performance in real-time applications
like object tracking and platform stabilization, the latter of which is described next.
One of the areas of interest in FSO is the use of fine angle pointing and tracking,
which would use micro positioning devices to control a laser beam. These allow for
precise angular control of an outgoing laser, being able to compensate for platform jitter

49

or other small and rapid disturbances. The disadvantage of these FPAT systems is that
their overall angular range is very small, on the order 1.5 [9]. So if the whole system
needs to compensate for more than that, the FPAT couldnt work. To mitigate this, a
FPAT unit can be mounted onto one of the gimbals found in this paper, also known as a
CPAT, or coarse angle pointing and tracking unit. The CPAT unit would then cancel out
the gross motion, allowing the FPAT to have a steady platform to work from. To achieve
this performance, a tilt sensor (or gyroscope) paired with a DSP board and the motors
encoders would calculate the required rotation to keep the platform level and then move
the motors to do so. In order to cancel out gross motion greater than 2 Hz (based from
preliminary experiments), a DSP approach would have to be used. The C6713
Evaluation Board from TI has the capabilities to do this, and is reasonably priced for
prototyping. For applications that require the use of only the RF radios, a much smaller
and lighter gimbal could be fabricated, which would reduce cost and increase reconnect
speeds. The author has actually begun development of these units, but they will not be
finished in time to be described in this thesis. They employ standard package DC
brushless servo motors from Parker Motion with 20,000ppr encoders, with the entire
gimbal only weighing about 10kg. Figure 4.1 below shows the conceptual design of the
unit.

50

Figure 4.1: Conceptual design of light-weight RF servo gimbal

There are many other modifications that can be made to these units (sliprings, clear
domed covers, honeycomb frame, etc), but the ones listed above are those that will cause
the most significant increases in performance. While these gimbals are only prototypes,
they will be very useful for proving concepts developed by other researchers. Directional
communications are fast becoming an important part of the broadband universe, and
these agile gimbals will open up new opportunities in reconfigurable networks.

51

Appendix A: Gimbal CAD Design Drawings

Figure A.1: Unloaded Gimbal Top View

52

Figure A.2: Gimbal side view

53

Figure A.3: Gimbal Front View

54

Bibliography
[1]

http://www.bodine-electric.com/etorq

[2]

http://www.web-books.com/eLibrary/Engineering/Circuits/AC/02459.png

[3]

www.orientalmotor.com/products/pdfs/C_VEXTA/AS-ASC_Series_Brochure.pdf

[4]

http://zone.ni.com/devzone/cda/tut/p/id/4623

[5]

Crowder, R. Electric Drives and Their Controls, Clarendon Press, 1983

[6]

Kaiser, D. Fundamentals of Servo Motion Control, Parker Compumotor, 2007

[7]

http://www.copleycontrols.com/motion/downloads/zip/XenusUserGuide.zip

[8]

PT Design, Feed-forward in Position-Velocity Loops, PT Design, 2000

[9]

Ho, T-H. Pointing, Acquisition, and Tracking Systems for Free-Space Optical

Communications Links, University of Maryland, 2007

[10]

http://support.motioneng.com/Downloads-Notes/Tuning/piv_tune_vel_loop.htm

[11]

PT Design, Cascaded Position-Velocity Loops, PT Design, 2000

[12]

http://www.rad.com

[13]

http://www.canobeam.com

[14]

http://www.o-navi.com/falcongx.htm

[15]

http://www.ece.umd.edu/mog/tracking.avi

55